You need to enable JavaScript to run this app.
优惠活动
大模型
产品
解决方案
定价
更多

求助:无NVIDIA GPU的MacOS系统下如何成功安装CuPy

Hey there, I’ve run into this exact issue before on macOS without an NVIDIA GPU—let’s walk through how to get CuPy’s CPU-only version installed properly.

First, prep your build environment

macOS requires some tools to compile CuPy from source (since pre-built CPU-only wheels aren’t always available for macOS). Start by installing Xcode Command Line Tools if you haven’t already:

xcode-select --install

Follow the prompts to complete the installation—this gives you the gcc/clang compilers needed for the build.

Option 1: Install via pip with CPU-only mode

The key here is to tell CuPy to skip CUDA detection entirely. Run this command, which sets an environment variable to disable CUDA checks during installation:

CUPY_NO_CUDA=1 pip install cupy

If you’re using a specific Python version (like your 3.7 env), make sure you’re using the pip associated with that environment (e.g., pip3.7 instead of pip if needed).

Option 2: Install from source (if pip install fails)

If the pip method still hits issues, compiling from source with the --no-cuda flag will work. You already cloned the repo earlier, so just adjust your install command:

git clone https://github.com/cupy/cupy.git
cd cupy
git submodule update --init
pip install -e . --no-cuda

The --no-cuda flag tells the setup script to ignore CUDA entirely and build only the CPU-compatible components.

Why your earlier attempts failed

All the commands you tried before didn’t explicitly disable CUDA detection. CuPy’s default behavior is to look for a CUDA environment first, and since macOS doesn’t have NVIDIA’s CUDA toolkit installed, it threw the "CUDA environment is invalid" error you saw. By setting CUPY_NO_CUDA=1 or using --no-cuda, we bypass that check entirely.

Verify the installation

Once installed, test it out in Python to make sure everything works:

import cupy as cp
test_array = cp.array([1, 2, 3])
print(test_array)

If you see array([1, 2, 3]) printed without errors, you’re good to go!

内容的提问来源于stack exchange,提问作者Nold

相关产品推荐
方舟 Agent Plan

超全模态模型 × Harness 升级,最新支持 Deepseek-V4.1-Flash、GLM-5.3 系列、Doubao-Seedream-5.0-pro、Kimi-K3 (部分), 限时 9.9 元起

最近更新时间:2026.04.30 21:07:36